Available online at www.sciencedirect.com
ScienceDirect
Procedia Computer Science 62 (2015) 343 – 351
The 2015 International Conference on Soft Computing and Software Engineering (SCSE 2015)
A Cloud to Mobile Application for Consumer Behavior Modification
Dustin Wrighta , Xing Yana , Pooja Srinivasa , Atieh Kashania , Yusuf Ozturka
a San
Diego State University, 5500 Campanile Dr., San Diego, CA 92129 USA
Abstract
Consumer behavior modification in the residential energy market aims to monitor consumer energy usage and enforce behavior
change through energy pricing or showing the impact to the individual consumer’s CO2 emissions when using energy at peak
hours versus at non-peak hours. This study presents a cloud connected mobile application and cloud based architecture that
focuses on improving the homeowner’s “know” and “care”, aiming to influence actions through transformation of moral in addition
to monetary savings. The system developed accesses consumer energy consumption using smart meters installed on customer
premises, predicts future energy consumption using a cloud application and finally provides the “know” and “care” information to
the user in a mobile gaming application. The system presents a cloud based solution for effecting customer behavior modification
in the electricity market with a mobile interface that encourages consumers to offset their energy usage during peak hours in order
to achieve full energy efficiency potential.
c 2015
Elsevier
2015The
TheAuthors.
Authors.Published
Published
©
by by
Elsevier
B.V.B.V.
This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of organizing committee of The 2015 International Conference on Soft Computing and Software
Peer-review
responsibility
of organizing committee of The 2015 International Conference on Soft Computing and Software
Engineeringunder
(SCSE
2015).
Engineering (SCSE 2015)
Keywords: Cloud computing, User behavior modification, Internet of things, Residential energy management
1. Introduction
In the current regulated energy transmission and distribution systems, electricity prices are generally averaged over
the entire year and the normal pricing is based on cost-of-service. This average electricity cost is used to charge
customers at a fixed rate, with real-time electricity pricing not being employed. The U.S. Energy Policy Act of 2005
suggests that utility companies should provide customers with time-based rates 1 , which can benefit both the utility
and the consumer. One of the primary purposes of implementing real-time pricing is consumer behavior modification
in which consumers are encouraged to offset demand during peak hours to off-peak hours, thus reducing the strain on
the grid. However, due to technical limitations on the demand side, the full implementation of time-based rates is still
considered a challenge 2 .
A method of effecting consumer behavior modification in energy consumption is demand response programs in
which consumers who reduce their usage during peak hours are rewarded. The U.S. Department of Energy reported
∗
Corresponding author. Tel.: +1-858-869-5579.
E-mail address: [email protected]
1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of organizing committee of The 2015 International Conference on Soft Computing and Software Engineering
(SCSE 2015)
doi:10.1016/j.procs.2015.08.412
344
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
in 2006 on the benefits of demand response in electricity markets and recommendations for achieving these benefits 3 .
Price based demand response programs, in which the customer is rewarded with reduced monetary expense for offsetting their demand during peak hours, can be effective programs for behavior modification. Such programs are divided
into time-of-use (TOU) programs 4,5,6,7 and real-time pricing (RTP) programs. Research on RTP programs has shown
viability in the consumer benefits through case studies, as well as preliminary models for achieving RTP 2,8,9 .
Energy prices can directly influence consumer demand and supplier generation 10,11,12 . Specifically, energy pricing
can be an effective control signal for imminent responses from the distributed and independent energy supplier community against instant energy demand alterations. However, it has been shown that using energy prices as a control
signal is only effective on a portion of the population 13 . In order to maximize the effectiveness of demand response
programs, one must provide more than monetary incentives as well as tailor the incentives to the preferences of the individual consumer. An incentive worth exploration is the environmental impact in terms of carbon dioxide emissions
of energy use during peak hours. In this, the combination of price based demand response and environmental impact
based demand response can influence actions of the consumer through the prospect of both monetary and moral savings. To accomplish this, a feedback system is required in which a third party monitor’s the consumer’s habits and
delivers specific control signals tailored to the consumer which improve both their “know” and “care”, making the
consumer more cognizant of their habits and influencing their desire to make changes which will help realize a more
energy efficient electricity market.
In this paper we propose a cloud based system architecture for enabling customer behavior modification in the
electricity market. Cloud computing is an efficient, scalable, and cost-effective technology which is well suited to
internet of things applications. The proposed architecture uses closed loop feedback to monitor customer energy use,
perform processing and calculation in the cloud, and provide an appropriate real-time pricing signal which will affect
customer behavior. The pricing signal influences the consumer with both monetary and carbon dioxide emission based
incentives. The goal of this architecture is to offset the total demand on the grid during peak hours to other periods
of the day in order to realize full energy efficiency potential. Energy usage is collected using smart meters installed
on customer premises and delivered to a third party server from the local utility using the Green Button application
programming interface (API). Using the historical demand, the future 72 hours of energy usage is predicted using
support vector regression, with which the optimal energy price is estimated. This price is then delivered to the
consumer using a series of web services which are consumed by a mobile application.
The rest of the paper is structured as follows. Section 2 presents the proposed system architecture and the various
software components involved in the system. Section 3 discusses the major interfaces required by the proposed
architecture. Section 4 discusses the mobile interface used in the system, and finally Section 5 gives a discussion of
the important aspects of the architecture and elaborates on future work.
2. Closed Loop Software Architecture
The software presented in this study aims to monitor and change a consumer’s energy consumption behaviors
which is driven not only by need but also by habits. The software developed aims to force behavioral change by
pricing the energy in such a way that users have an incentive to move their energy consumption to off-peak hours.
The software also aims to create guilt in the energy consumer by showing his/her individual CO2 emissions and
encourages them by providing information related to how much he could reduce his CO2 emissions by changing his
energy consumption behaviors. Since the system is designed as a voluntary participation solution, it is expected to
start with a small number of participants and add more consumers in the future. This requires the software solution
presented to be scalable with more resources added as new consumers join the system.
2.1. Cloud Based Infrastructure
The system is implemented in the cloud in order to maximize the scalability of the system. A high level block
diagram of the system developed is presented in Figure 1. Meter data is first collected from customer premises using
smart meters. The smart meter may be read directly using a ZigBee 14 gateway or indirectly via the energy utility. This
architecture supports both scenarios, with direct data being received using a Rainforest EAGLE device and indirect
acquisition relying on the local utility company (SDG&E). In the latter case, meter readings are stored in a database
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
Fig. 1. Data Pipeline
owned by the utility company and made available to third parties through an interface called Green Button Connect.
This program offers third party energy management agents the consumers energy consumption with their consent
through the Green Button API.
Using the Green Button API, a file with hourly consumption data for all customers registered with the third party
is delivered to the server of the third party entity. The utility is provided an account on the third party server which
is used for file transfers. Using the Secure File Transfer Protocol (SFTP), the utility sends files to the third party
server with new subscription information and historical energy consumption. The decision to use SFTP as opposed to
web services was made by the utility prior to this study and is a common interface for all of the third party entities.
Subscription files provide identifying information about users who have recently registered to use our application
through Green Button, including their email and a unique key. Historical demand is provided for up to 13 months with
15 to 60 minutes of resolution. Our architecture uses a datastore to maintain all of the data received and calculated.
This datastore consists of a relational database and local file storage on the server. A relational database is chosen as
opposed to a NoSQL database due to the well-defined schemas for all of the data in the system. In addition, AWS
provides a relational database service (RDS) which allows for the creation and scaling of relational databases in the
cloud. For this cloud based system, Amazon EC2 is chosen as the platform for hosting our server, where interation
with the utility is achieved through Green button API and interaction with consumers is achieved through web services
developed using the Spring Framework.
Our approach to enabling behavior modification is to predict the demand on the grid for the future 72 hours, and
based on this come up with an hourly pricing signal which will offset the demand during the peak hours. This involves
storing the hourly data sent through Green Button, using support vector regression to predict future demand for each
individual customer that has registered with us, and calculating the optimal price. For details on the support vector
regression implementation used see 15 , and for details on the price estimation formulations refer to 16 . To accomplish
all of the required tasks on the server, we have developed an infrastructure which coordinates the collection, storage,
analysis, and presentation of all data involved in the system. This cloud infrastructure is depicted in detail in Figure 2.
In this study, San Diego Gas and Electric (SDG&E), the local utility company, is provided with a user account on
the EC2 server so that historical and daily energy consumption records can be transferred using SFTP. This interface
provides energy consumption data for each individual user with 15 to 60 minutes of resolution. Historical energy
consumption data is available from SDG&E for 13 months, with new data being provided once every 24 hours for
the previous day. When a new energy consumption file is sent to the server, a demand recording service is executed
as a background task to extract all energy demand data and metadata associated with the meter readings. This data is
subsequently stored in a MySQL database which is part of our datastore.
Forecasting is performed after storing energy consumption data in the database every 24 hours as new data becomes
available. The first step is to acquire all necessary forecasting inputs and format them as necessary. The inputs gathered
include hourly electricity consumption, hour of the day, indication of the day (weekday or weekend and holiday),
and working hours. Other relevant information such as weather data and solar radiation will be considered in the
future. Once the input has been prepared, the developed demand forecasting algorithm using support vector machines
is executed. The demand forecasting algorithm is implemented using the open source GNU Octave mathematical
345
346
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
Fig. 2. Proposed Cloud Infrastructure
Fig. 3. Pricing Signal Generation
computation package. For each user, a forecast file is generated with 3 days of hourly energy usage and 14 days of
average energy usage, which is stored in a cache on the servers local file system.
Next, a 3-tier price is determined for the 72 hours following the most recently recorded demand. In addition to
forecasted electricity demand, real-time market and day-ahead market electricity prices from California ISO (CAISO)
are used as inputs as shown in Figure 1. Prices are stored starting from January 1, 2014, and the latest prices are
maintained at all times and collected asynchronously from the price estimation algorithm. Price estimation begins by
aggregating the forecasted demand from the cache. Real-time market prices from the previous 7 days are extracted
from the data and day-ahead market prices are extracted for the previous 7 days and up to 3 days in the future.
Depending on the availability of the data, some hours or days may use prices from the previous hour or day. Once the
inputs are formatted, price estimation is performed and the results are stored locally.
A block diagram of the generation of the individual consumer’s pricing signals is shown in Figure 3. The price
signal for each individual user is dependent on the average (μi ) and standard deviation (σi ) of the demand for the
whole grid at each hour (i). If the predicted demand for the user Pi is such that Pi < μi − σi , the user is placed in
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
tier one. In the same way, if Pi > μi + σi they are placed in tier three; otherwise, they are placed in tier two. This
causes the price a user receives to reflect their contribution to the total demand on the grid. In addition to price, we
are interested in using carbon dioxide emissions as a control signal. The transformation from electric energy (kWh)
to carbon dioxide emissions (lbs of CO2 ) is based on a conversion found by the EPA 17 . The conversion uses the type
of generation (i.e. coal vs. natural gas) and time of day. In our system, we use the tier placement found for price
to represent different types of generation, causing the amount of carbon dioxide emissions to reflect the individual
customer’s impact on the total demand. Future work will include the percentage of generation due to different fuel
types at each hour to determine a highly accurate estimate of the amount of CO2 emitted due to the individual user’s
energy consumption.
The proposed system provides a mobile application for customers to view their historical and forecasted demand,
as well as historical and estimated energy prices. To access this data, a simple representational state transfer (REST)
interface that allows the mobile application to request data from our datastore is implemented. The Spring Framework
is used to develop several RESTful web services that the mobile application utilizes to make GET and POST requests
to the server. Consumers can then view what their expected energy usage will be for the day and based on the estimated
cost can alter their behavior. This in turn, acts as a real-time TOU pricing feedback mechanism for the residential
energy customers, making this a closed-loop system.
2.2. Software Components
The software for monitoring and encouraging customer behaviour modification is a complex networked software
designed with scalability and fault tolerance in mind. The following software frameworks are employed in engineering
the software presented in this paper.
2.2.1. Amazon Web Services
Amazon Web Services (AWS) is a cloud platform for clients to base their web infrastructure. AWS provides many
of the components one needs for an entire IT services backend with pay-as-you-go pricing. In this, one is able to
pick and choose IT services individually and at any scale required for the specific use cases with a monthly fee. The
primary services used in our system are Elastic Compute Cloud (EC2) and the Relational Database Service (RDS).
EC2 is a web service which provides the customer with scalable computing resources housed on Amazon’s servers.
We are using an Ubuntu 14.04 server instance at the m3.medium tier, which includes an Intel Xeon E5-2670 v2 (ivy
bridge, 2.5 GHz) processor with 4 GB of RAM and 30GB of storage. The RDS is a web service which allows one to
host relational databases in the cloud with a variety of options in regards to size and performace. We host one MySQL
database with 20GB of storage at the db.t1.micro tier.
2.2.2. Spring Framework
The Spring Framework is a comprehensive programming and configuration model for Java-based enterprise applications 18 . Spring provides numerous components for achieving different tasks in a Java-based web application,
with features such as dependency injection, aspect-oriented programming, model-view-controller (MVC) application
infrastructure, and RESTful (representational state transfer) web service framework. The proposed architecture relies
heavily on the Spring RESTful framework for implementing web services. Spring can be easily integrated into a
Java EE project by using Maven, and makes the development of web services simple in that it abstracts much of the
backend code. Our web application which utilizes the Spring Framework is hosted on our Amazon EC2 server.
2.2.3. Octave and LibSVM
Octave is a mathematical computation package provided freely under the GNU General Public License. The
package provides an interpreted language very similar to Matlab, and therefore can run most Matlab scripts with
minimal modification. LibSVM 15 is a widely used library for support vector machines which is available in multiple
programming languages. The library has been actively developed since 2001. In our proposed architecture, the
LibSVM implementation of support vector regression is used to solve for the predicted energy demand of individual
consumers. This is ran as a Matlab script using Octave which is installed on the server. In addition, Octave is used to
determine the optimal price of energy which will shift demand off of the peak hours.
347
348
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
Fig. 4. Utility-Server Interaction
3. System Interfaces
The software architecture presented in this paper spans several hardware and sofwtare platforms thus requires well
defined interfaces between components of the system developed on different platforms or context. Two main interfaces
are very critical in design of the system thus will be discussed below.
3.1. Utility-Server
The first interface in the system is between the utility and the third party server via the Green Button API. The
interaction between the utility and server is shown in Figure 4.
A third party must apply to become officially a part of the utility’s “Green Button Connect My Data” program.
During this process, the third party is reviewed to ensure they meet certain business and security requirements as
deemed appropriate by the utility. Once approved, customers may release their data to this third party through Green
Button. To accomplish this, they must first consent to release their meter readings to the third party. This is made
possible through the utility’s web portal. The customer follows a simple prompt in their account on the utility’s web
portal where they select a third paty to which they consent to release their data. All third parties which have signed up
with the utility are displayed as options for the customer to select. Once the customer has volunteered their data, the
utility sends the subscription and historical consumption files to the third party server for processing. Each day, the
utility then sends consumption data for each customer for the previous 24 hours. This data is subsequently added to
the server’s datastore. The customer may opt out of this service at any time through the utility’s web portal.
3.2. Server-Customer
The next interface, between the server and the customer, is a series of web services hosted on the server which
deliver consumption and pricing data to a mobile application. The customer must download the mobile app from the
Google Play store, at which time their identity is verified by sending a code to the email address they used to sign up
for Green Button and having them enter that code. In addition, each customer is given a unique key when they sign
up for Green Button, which is used to identify all of the customer’s data. Using this key and the customer’s email
address, the mobile application makes several GET requests to retrieve historical and predicted energy demand, as
well as the historical and predicted price of energy. In addition, we use this interface to give the customer suggestions
on the optimal time to use their appliances that will save them the most money. The mobile application also makes
POST requests to update the server on their decisions in terms of shifting appliance usage so that we can keep track
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
Fig. 5. Consumer Interface
of their savings. These savings are compared to other customers in the system and presented to the user so that they
can have a benchmark for comparison.
Since we are also interested in carbon dioxide emissions based demand response, the mobile application gives the
user the option to view their cost and savings in terms of pounds of carbon dioxide emitted. The mobile application
uses the same web services to retrieve carbon dioxide emissions, simply changing a parameter in the request URL.
This gives us feedback from the user in terms of their response to the environmental impact control signal.
4. Customer Behavior Modification Interface
The focus of the customer behavior modification interface is to improve the homeowners “know” and “care”,
aiming to influence actions through transformation of moral in addition to monetary savings. Behavior change is
enforced through energy pricing and showing the impact to the individual consumers CO2 emissions when using
energy at peak hours versus at non-peak hours. We have chosen the mobile phone platform as the consumer interface
and developed an Android application to improve the user’s “know” and “care”. The application developed, in addition
to providing the user information about their energy consumption and their contribution to CO2 levels, offers a game
environment where users compete to reduce their CO2 emissions and increase monetary savings. Each user is provided
the option to create a social network and invite their friends to join their social network. Users are then provided
information on their social network and how they stack up against them. Figure 5 provides sample user interface
screens designed for customer behavior modification.
Security of the energy data is provided through a two level authentication process where user’s email is used as a
mean to provide the user a key to enable access to his/her energy data.
5. Discussion
In this paper, we propose a cloud based architecture to enable customer behavior modification in the electricity
market. The solution we have developed is a comprehensive end-to-end system with closed loop consumer feedback.
Energy demand gathered from individual customers is used to predict the load on the grid and set an optimal price of
energy which will offset demand during peak periods. The demand is predicted using support vector regression with
historical demand, time of day, and weekday or weekend/holiday used as the features. The optimal price is calculated
349
350
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
Fig. 6. Network Traffic for Jan. 16 - 24
using the aggregated demand for the entire grid and the day ahead and real time market prices. In addition to price,
environmental impact is used as a control signal to encourage consumers to shift their demand in order to maximize
the number of users influenced. A mobile application serves as a user interface which retrieves data stored on the
server and presents the individual customer’s demand and price of electricity. Finally, the mobile application helps
the consumer determine the best time to use their appliances in order to maximize their savings.
The developed system is very accessible to consumers, with a two step process required to start using the mobile
application. Consumer’s need only to register for Green Button through the utility company and then download the
mobile application and log in through our interface. After the server is sent their consumption data, we are able to
forecast their future demand and calculate their related cost of consumption and CO2 emissions. The consumer is able
to view these figures using the mobile application within 24 hours of their registration with Green Button.
Our initial test environment has consisted of between 2 and 24 users at a time. Firgure 6 shows network traffic
from January 16th to the 24th. This graph indicates that we experience relatively low traffic during most periods, with
spikes at midnight of each day when pricing and forecasting are executed. Given that this architecture is intended
to support behavior modification for a large subpopulation, scalability is a critical consideration as the number of
subscribers to the system increases. AWS easily enables one to scale up processing and storage needs by providing a
tiered pricing structure based on the processing requirements. Using EC2, one can save images of systems currently
in use and bring up new systems with greater processing power using these saved images. We utilize this in order to
continually upgrade our backend as the need arises with relatively low cost overhead. In this, the proposed system can
support a large body of users in a simple and cost effective way.
In addition to pricing signals and carbon dioxide emission signals to modify customer behavior, we have implemented a momentary assessment tool as a part of the mobile-server interface which allows us to obtain extra feedback
on the effectiveness of our system. With this interface we are currently providing energy saving tips as well as tips
about the optimal time to use appliances during the day given the predicted prices. When the user observes these tips,
we ask them a short yes or no question on whether they found the tip helpful or if they would follow our suggestion.
Using this, we can start to develop a metric for the effectiveness of the customer behavior modification.
Our current direction is to expand the number of factors used to effect customer behavior modification and identify
metrics for the influence of these factors in customer behavior. This includes social, cultural, and behavioral factors
will aid in achieving full energy efficiency potential. Our specific objectives are to use data collected in the field
and data related to consumer interactions with the utility to study the role of society, culture, and habits in energy
efficiency uptake, and identify specific product/service marketing techniques that may influence these factors, as well
as collect data that will enable identification of specific metrics for social, cultural, and behavioral phenomena that
can be projected forward 5 to 10 years, and that may be included within improved energy demand forecasting models
Dustin Wright et al. / Procedia Computer Science 62 (2015) 343 – 351
and energy efficiency studies. In order to accomplish this we will first augment the number of users in our system and
target specific groups of users which have desired social and cultural backgrounds for study.
Acknowledgments
This work has been supported in part by California Institute of Energy and Environment (Grant No: 500-01-043),
Regents of University of California (CIEE Sub award Number: PODR05-X22), and San Diego Gas and Electric
(SDG&E).
References
1. Y. Ozturk, D. Senthilkumar, S. Kumar, G. Lee, An intelligent home energy management system to improve demand response, Smart Grid,
IEEE Transactions on 4 (2) (2013) 694–701.
2. Q. Zhang, X. Wang, M. Fu, Optimal implementation strategies for critical peak pricing, in: Energy Market, 2009. EEM 2009. 6th International
Conference on the European, IEEE, 2009, pp. 1–6.
3. Q. QDR, Benefits of demand response in electricity markets and recommendations for achieving them.
4. H. Aalami, G. Yousefi, M. P. Moghadam, Demand response model considering edrp and tou programs, in: Transmission and Distribution
Conference and Exposition, 2008. T&# x00026; D. IEEE/PES, IEEE, 2008, pp. 1–6.
5. H. Aalami, M. Moghadam, G. Yousefi, Optimum time of use program proposal for iranian power systems, in: Electric Power and Energy
Conversion Systems, 2009. EPECS’09. International Conference on, IEEE, 2009, pp. 1–6.
6. S. Zeng, J. Li, Y. Ren, Research of time-of-use electricity pricing models in china: A survey, in: Industrial Engineering and Engineering
Management, 2008. IEEM 2008. IEEE International Conference on, IEEE, 2008, pp. 2191–2195.
7. Y. Tang, H. Song, F. Hu, Y. Zou, Investigation on tou pricing principles, in: Transmission and Distribution Conference and Exhibition: Asia
and Pacific, 2005 IEEE/PES, IEEE, 2005, pp. 1–9.
8. J. Y. Choi, S.-H. Rim, J.-K. Park, Optimal real time pricing of real and reactive powers, Power Systems, IEEE Transactions on 13 (4) (1998)
1226–1231.
9. A. J. Conejo, J. M. Morales, L. Baringo, Real-time demand response model, Smart Grid, IEEE Transactions on 1 (3) (2010) 236–242.
10. P. Finn, C. Fitzpatrick, D. Connolly, Demand side management of electric car charging: Benefits for consumer and grid, Energy 42 (1) (2012)
358–363.
11. A.-H. Mohsenian-Rad, A. Leon-Garcia, Optimal residential load control with price prediction in real-time electricity pricing environments,
Smart Grid, IEEE Transactions on 1 (2) (2010) 120–133.
12. A.-H. Mohsenian-Rad, V. W. Wong, J. Jatskevich, R. Schober, A. Leon-Garcia, Autonomous demand-side management based on gametheoretic energy consumption scheduling for the future smart grid, Smart Grid, IEEE Transactions on 1 (3) (2010) 320–331.
13. T. Ida, K. Ito, M. Tanaka, Using dynamic electricity pricing to address energy crises: Evidence from randomized field experiments, Tech.
rep., Mimeo (2013).
14. Z. Alliance, Zigbee specification (2006).
15. C.-C. Chang, C.-J. Lin, Libsvm: a library for support vector machines, ACM Transactions on Intelligent Systems and Technology (TIST)
2 (3) (2011) 27.
16. S. Datchanamoorthy, S. Kumar, Y. Ozturk, G. Lee, Optimal time-of-use pricing for residential load control, in: Smart Grid Communications
(SmartGridComm), 2011 IEEE International Conference on, IEEE, 2011, pp. 375–380.
17. How much carbon dioxide is produced per kilowatthour when generating electricity with fossil fuels?, http://www.eia.gov/tools/
faqs/faq.cfm?id=74&t=11, accessed: 2014-11-22.
18. Spring framework overview, http://projects.spring.io/spring-framework/, accessed: 2014-11-22.
351